Comparative Analysis of CNN and LSTM Neural Network Models: Influencing Factors in Agricultural Machinery Fault Detection

Abstract

Predictive maintenance is increasingly vital for agricultural machinery as sensor-based monitoring becomes more prevalent. Researchers are investigating reliable fault detection across diverse operating conditions and real-world constraints. This study compares three deep learning approaches for fault detection, maintenance, and repair planning using literature data, and evaluates convolutional, sequential-based, and hybrid architectures within a unified framework.. A Convolutional Neural Network (CNN) is an artificial neural network model used specifically for processing data with a grid-like topology, such as images or time-series signals. Long Short-Term Memory (LSTM) network is a special type of Recurrent Neural Network (RNN) designed to learn long-term dependencies in sequential data. The CNN and LSTM models were evaluated for accuracy, efficiency, robustness to noise, and interpretability. Results showed that convolutional models excelled when fault patterns were localized and easier to identify in transformed formats such as the frequency domain or images. These models also provided fast inference with minimal resource use. Sequence models performed well, gradually degrading as they leveraged longer temporal contexts to detect progressive changes. In environments with multiple sensors and mixed data types, hybrid architectures yielded the most consistent results by combining spatial feature extraction with temporal modeling, enhancing accuracy and suitability for resource-limited settings. The best accuracy percentages for the effective PdM design parameters are 91.2% for the CNN model, 93.8% for the LSTM model, and 96.4% for the hybrid CNN+LSTM model. Similarly, the dropout rates are 0.1–0.4 for CNN, 0.2–0.5 for LSTM, and 0.2–0.5 for the CNN+LSTM model. Overall, model selection should depend on the data structure and monitoring goals, with hybrid models offering a balanced solution for detecting short-term anomalies and long-term deterioration.